423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-03 04:02:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

update readme

Browse Source 
Former-commit-id: e4eec9ddfd3a9688733e018a96274dff0d5d9962
This commit is contained in:
hiyouga 2023-08-18 01:51:55 +08:00
parent 429a9e2512
commit e93e9641f5
2 changed files with 78 additions and 73 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

77
README.md
View File

@ -164,7 +164,9 @@ We strongly recommend using the all-in-one Web UI for newcomers since it can als
Currently the web UI only supports training on **a single GPU**. Currently the web UI only supports training on **a single GPU**.
### Pre-Training ### Train on a single GPU
#### Pre-Training
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
@ -187,7 +189,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--fp16 --fp16
``` ```
### Supervised Fine-Tuning #### Supervised Fine-Tuning
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
@ -210,7 +212,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--fp16 --fp16
``` ```
### Reward Modeling #### Reward Modeling
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
@ -234,7 +236,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--fp16 --fp16
``` ```
### PPO Training #### PPO Training
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
@ -255,10 +257,11 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--save_steps 1000 \ --save_steps 1000 \
--learning_rate 1e-5 \ --learning_rate 1e-5 \
--num_train_epochs 1.0 \ --num_train_epochs 1.0 \
--plot_loss --plot_loss \
--fp16
``` ```
### DPO Training #### DPO Training
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
@ -357,40 +360,15 @@ deepspeed --num_gpus 8 --master_port=9901 src/train_bash.py \
</details> </details>
### Evaluation (BLEU and ROUGE_CHINESE) ### Export model
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ python src/export_model.py \
--stage sft \
--model_name_or_path path_to_your_model \ --model_name_or_path path_to_your_model \
--do_eval \
--dataset alpaca_gpt4_en \
--template default \ --template default \
--finetuning_type lora \ --finetuning_type lora \
--checkpoint_dir path_to_checkpoint \ --checkpoint_dir path_to_checkpoint \
--output_dir path_to_eval_result \ --output_dir path_to_export
--per_device_eval_batch_size 8 \
--max_samples 100 \
--predict_with_generate
```
We recommend using `--per_device_eval_batch_size=1` and `--max_target_length 128` at 4/8-bit evaluation.
### Predict
```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--stage sft \
--model_name_or_path path_to_your_model \
--do_predict \
--dataset alpaca_gpt4_en \
--template default \
--finetuning_type lora \
--checkpoint_dir path_to_checkpoint \
--output_dir path_to_predict_result \
--per_device_eval_batch_size 8 \
--max_samples 100 \
--predict_with_generate
``` ```
### API Demo ### API Demo
@ -425,15 +403,40 @@ python src/web_demo.py \
--checkpoint_dir path_to_checkpoint --checkpoint_dir path_to_checkpoint
``` ```
### Export model ### Evaluation (BLEU and ROUGE_CHINESE)
```bash ```bash
python src/export_model.py \ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--stage sft \
--model_name_or_path path_to_your_model \ --model_name_or_path path_to_your_model \
--do_eval \
--dataset alpaca_gpt4_en \
--template default \ --template default \
--finetuning_type lora \ --finetuning_type lora \
--checkpoint_dir path_to_checkpoint \ --checkpoint_dir path_to_checkpoint \
--output_dir path_to_export --output_dir path_to_eval_result \
--per_device_eval_batch_size 8 \
--max_samples 100 \
--predict_with_generate
```
We recommend using `--per_device_eval_batch_size=1` and `--max_target_length 128` at 4/8-bit evaluation.
### Predict
```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--stage sft \
--model_name_or_path path_to_your_model \
--do_predict \
--dataset alpaca_gpt4_en \
--template default \
--finetuning_type lora \
--checkpoint_dir path_to_checkpoint \
--output_dir path_to_predict_result \
--per_device_eval_batch_size 8 \
--max_samples 100 \
--predict_with_generate
``` ```
## TODO ## TODO

74
README_zh.md
View File

@ -164,7 +164,9 @@ CUDA_VISIBLE_DEVICES=0 python src/train_web.py
目前网页 UI 仅支持**单卡训练**。 目前网页 UI 仅支持**单卡训练**。
### 预训练 ### 单 GPU 训练
#### 预训练
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
@ -187,7 +189,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--fp16 --fp16
``` ```
### 指令监督微调 #### 指令监督微调
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
@ -210,7 +212,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--fp16 --fp16
``` ```
### 奖励模型训练 #### 奖励模型训练
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
@ -234,7 +236,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--fp16 --fp16
``` ```
### PPO 训练 #### PPO 训练
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
@ -258,7 +260,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--plot_loss --plot_loss
``` ```
### DPO 训练 #### DPO 训练
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
@ -357,40 +359,15 @@ deepspeed --num_gpus 8 --master_port=9901 src/train_bash.py \
</details> </details>
### 指标评估（BLEU 分数和汉语 ROUGE 分数） ### 导出微调后的模型
```bash ```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \ python src/export_model.py \
--stage sft \
--model_name_or_path path_to_your_model \ --model_name_or_path path_to_your_model \
--do_eval \
--dataset alpaca_gpt4_zh \
--template default \ --template default \
--finetuning_type lora \ --finetuning_type lora \
--checkpoint_dir path_to_checkpoint \ --checkpoint_dir path_to_checkpoint \
--output_dir path_to_eval_result \ --output_dir path_to_export
--per_device_eval_batch_size 8 \
--max_samples 100 \
--predict_with_generate
```
我们建议在量化模型的评估中使用 `--per_device_eval_batch_size=1` 和 `--max_target_length 128`。
### 模型预测
```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--stage sft \
--model_name_or_path path_to_your_model \
--do_predict \
--dataset alpaca_gpt4_zh \
--template default \
--finetuning_type lora \
--checkpoint_dir path_to_checkpoint \
--output_dir path_to_predict_result \
--per_device_eval_batch_size 8 \
--max_samples 100 \
--predict_with_generate
``` ```
### API 服务 ### API 服务
@ -425,15 +402,40 @@ python src/web_demo.py \
--checkpoint_dir path_to_checkpoint --checkpoint_dir path_to_checkpoint
``` ```
### 导出微调模型 ### 指标评估（BLEU 分数和汉语 ROUGE 分数）
```bash ```bash
python src/export_model.py \ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--stage sft \
--model_name_or_path path_to_your_model \ --model_name_or_path path_to_your_model \
--do_eval \
--dataset alpaca_gpt4_zh \
--template default \ --template default \
--finetuning_type lora \ --finetuning_type lora \
--checkpoint_dir path_to_checkpoint \ --checkpoint_dir path_to_checkpoint \
--output_dir path_to_export --output_dir path_to_eval_result \
--per_device_eval_batch_size 8 \
--max_samples 100 \
--predict_with_generate
```
我们建议在量化模型的评估中使用 `--per_device_eval_batch_size=1` 和 `--max_target_length 128`。
### 模型预测
```bash
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--stage sft \
--model_name_or_path path_to_your_model \
--do_predict \
--dataset alpaca_gpt4_zh \
--template default \
--finetuning_type lora \
--checkpoint_dir path_to_checkpoint \
--output_dir path_to_predict_result \
--per_device_eval_batch_size 8 \
--max_samples 100 \
--predict_with_generate
``` ```
## TODO ## TODO